A machine learning algorithm was developed for efficiently predicting the 3D (three-dimensional) spatiotemporal evolution process of tidal currents and analyzing their spatial distribution characteristics. In the algorithm, an extremely simplified multi-layer perceptron architecture, an embedded spatial information learning method, and a splicing-sharing method for tidal currents at different water depths were used to achieve a high-coverage, comprehensive, and systematic 3D tidal current prediction of the study area. The developed algorithm can efficiently predict the future time series of three-dimensional tidal current movement and solves the problem that existing algorithms are unable to analyze the similarity of the three-dimensional spatiotemporal distribution of tidal currents over many years. In this study, 3D tidal current evolutions in the southern waters of Liaoning Province, China, were analyzed. The Finite-Volume Coastal Ocean Model ocean model was used to simulate tidal currents in the study zone, generating a dataset to train the developed machine learning model. The trained model was then used to predict and analyze tidal currents. The prediction results show that the developed machine learning model has high prediction accuracy for tidal currents over a future period of 12 h, with R2 (R-Square) of 0.871, mean absolute error of 0.047 m/s and root mean square error of 0.152 m/s. Additionally, the developed machine learning model could effectively analyze the correlation of spatial distribution characteristics of tidal currents at different water depths, and tidal currents with similar evolution processes at different zones could also be classified.
Read full abstract